docs(skill/scitexification): future enhancement — tag-filtering for subset companion loads

[ywatanabe1989]

Future enhancement surfaced during PR #118 review.

Currently 00_playbook.md and 06_scitexification/SKILL.md declare four requires: companion skills — scitex-session, scitex-io, figrecipe, scitex-clew. A SAC agent that loads scitexification therefore pulls all four full SKILL.md trees into context. Measured aggregate: ~30K tokens (session 3.6K + io 11.3K + plt 2.1K + figrecipe 3.6K + clew 9.4K) — comfortably within the 256K window, so the simple full-load is the right default today.

A future tag-filtering mechanism could let downstream agents declare a narrower subset — e.g. "this capsule only does stages 1+2, give me just session + io" — using sub-tags like scitexification.io / scitexification.session already sketched in the umbrella SKILL.md. The narrower load would surface only the relevant subset of each companion skill (e.g. just scitex-io's save/load surface, not its mcp / scholar integration).

This is deferred, not blocking — the full ~30K load works for all current consumers (paper-scitex-clew, future scitexification users), and tag-filtering only becomes worth it when context budgets tighten or when a consumer is doing a strict stages-1+2 subset.

Action items if/when picked up:

• Extend scitex-dev skills CLI with a tags-expand --filter <sub-tag> mode that resolves scitexification.io to just the relevant scitex-io chapters.
• Document the sub-tag-to-companion-subset mapping in 00_playbook.md's "Required companion skills" section.
• Wire SAC spec.skills.required to honour sub-tags (no-op for SAC versions that don't yet support it).

Suggested labels: enhancement, skills, low-priority.